I’ve been following Meta’s workforce management practices, and the lawsuit from 26 former employees reveals something troubling about how we’re deploying AI in high-stakes decisions. The claim is straightforward but serious: Meta used internal AI tools to rank and score employees for layoffs in May 2024, but the system didn’t account for workers on protected medical or parental leave. That’s not just a technical oversight, it’s a legal and ethical failure.
The specifics matter here. According to the lawsuit, Meta leveraged a constellation of AI systems including Metamate (an internal AI assistant), employee-trained AI agents, and internal dashboards tracking AI token usage to determine termination lists for the company’s 10 percent workforce reduction. The problem? These tools scored and ranked workers without excluding those exercising their legal right to protected leave. The result was predictable: people on leave were disproportionately targeted for dismissal.
The Technical Flaw Behind the Human Impact
This case illustrates a fundamental issue I worry about constantly in the AI ethics space: we build systems optimized for one goal, deploy them in complex human contexts, and assume social and legal constraints will somehow emerge naturally. They don’t.
Meta’s AI tools were presumably trained on performance metrics, engagement data, and productivity signals. That makes sense for a ranking system in isolation. But the moment you apply that ranking to determine who gets fired, you’ve entered legal and moral territory that requires explicit constraints. A worker on leave doesn’t perform at their usual level not because they’re unproductive, but because they’re not working. The AI couldn’t distinguish between legitimate absence and poor performance because nobody told it to.
This is a implementation problem, not a fundamental limitation of the technology. You could absolutely build an AI system that ranks employee performance while properly excluding protected leave periods. Meta presumably has teams smart enough to do this. The question is whether anyone raised the issue before the system went live, and if they did, whether they were heard.
What This Means for Enterprise AI Deployment
I think there’s a broader lesson here for anyone building AI systems in corporate environments. When your system makes decisions that affect people’s livelihoods, income, or legal rights, you need to bake compliance into the architecture, not tack it on afterward. This isn’t just good ethics, it’s good engineering.
Meta’s statement that “workforce management decisions were made by people, not AI” misses the point entirely. Sure, humans made the final calls, but they did so using scores and rankings generated by AI that was missing critical constraints. That’s not absolution, that’s abdication of responsibility. The humans didn’t review the algorithm’s limitations before using its outputs to fire thousands of people.
For developers building machine learning systems in employment contexts, here’s the uncomfortable truth: your data rarely captures everything that matters. An employee’s performance during a quarter when they’re on parental leave isn’t a signal about their value or ability. It’s a signal about parental leave existing. You need domain expertise, legal review, and ethical scrutiny built into your process before you deploy, not after lawsuits force accountability.
The Precedent This Sets
What concerns me most is that this lawsuit, if it succeeds, might become a template for other similar cases. How many other companies have used AI to optimize workforce reduction without properly considering legal constraints? I’d wager Meta isn’t alone.
The irony is that properly implemented AI could actually make these decisions fairer and more consistent than purely human judgment. But that requires transparency, auditability, and a willingness to constrain the optimization to respect legal rights and human dignity. It requires asking hard questions during development, not during depositions.
Meta built tools to help make workforce decisions faster and at scale. That ambition isn’t wrong. But scaling human decision-making without scaling the ethical considerations behind it produces exactly what we see here: systems that inadvertently break the law while technically functioning as designed.
The real question isn’t whether AI should help with workforce management, but whether we’re willing to build AI systems that make hard, constraint-respecting choices when the simpler path is to ignore what we don’t measure.